Skip to main content
The Molpha MCP server lets an AI agent derive source IDs, run signing rounds, pay for them, publish results to Solana and build verifier calls for EVM and Starknet, all as structured tools instead of prompt glue. It speaks the Model Context Protocol and works with Claude, Cursor, VS Code, Codex and other MCP clients.

Choose an integration

Both run the same server code and return the same signed artifacts.

Tools at a glance

Inputs, outputs and error codes are in the tool reference.

How it fits together

The server is an adapter and a policy boundary, not a source of truth. Every round tool returns the signed attestation plus ready-made verifier arguments for each chain you ask for. Treat that attestation, not the agent’s summary of it, as the trust anchor.

Safety model

  • Every round spends. There is no idempotency key: a repeated call after an unclear error is a new round. Read state (describe_feed, get_x402_status) before retrying.
  • Limits live outside the prompt. On the hosted server, set a spending policy on your wallet. On a local server, use the dry-run lock and caps, which an agent cannot switch off from a conversation.
  • Untrusted payment terms are checked. Before anything is signed, the server checks a gateway’s 402 terms against what it derives itself: the protocol treasury as payee, the protocol USDC mint, the protocol price and the cluster.
  • Verify before acting. build_verifier_calldata builds arguments only. Your contract or service calls verify().